feat: QUIC Address discovery extension #2043

divagant-martian · 2024-11-14T16:13:32Z

Implements the Quic Address discovery extension IETF draft

Transport parameters are implemented with tests and 0rtt verification as spec'd
The spec declares two frames, one for informing of IpV4 addresses and one for informing IpV6 frames. This is implemented as a single structure with tests.
Transmission is done after the initial path has been established, when path validation is required, and sent with path responses. The last part is not required but provides a small reconfirmation of the observed external address to the peer, and it's allowed by the spec being observed address frames defined as probing frames. This is implemented with tests.
Retransmission is done by sending a fresh frame if it's identified as lost. This ensures that the path over which the frame is sent is always the correct one, i.e. That the reported observed address is always sent using the path for which the address was observed. This is implemented with tests.
Surfacing is done from the protocol implementation via Connection events.
Configuration is done with two methods to independently turn on or off the receiving and sending of reports.

Ralith

Thanks for working on this!

quinn/src/connection.rs

Ralith · 2024-11-22T20:26:06Z

quinn/src/connection.rs

@@ -636,6 +636,12 @@ impl Connection {
        // May need to send MAX_STREAMS to make progress
        conn.wake();
    }
+
+    /// Track changed on our external address as reported by the peer.
+    pub fn observed_external_addr(&self) -> watch::Receiver<Option<SocketAddr>> {


Explicitly document when this is None, whether it might contain spurious updates, and the influence of configuration and peer support.

Not sure about exposing the watch::Receiver as part of the public API, that seems like it's leaking a bunch of implementation details?

quinn/Cargo.toml

quinn-proto/src/tests/mod.rs

Ralith · 2024-11-22T20:34:44Z

quinn-proto/src/tests/mod.rs

+/// NOTE: this test is the same as zero_rtt_happypath, changing client transport parameters on
+/// resumption.
+#[test]
+fn address_discovery_zero_rtt_accepted() {


What additional coverage does this case supply? I don't think changing client transport parameters are ever a concern for 0-RTT.

quinn-proto/src/frame.rs

Ralith · 2024-11-22T21:29:15Z

quinn-proto/src/connection/mod.rs

@@ -3760,6 +3874,8 @@ pub enum Event {
    DatagramReceived,
    /// One or more application datagrams have been sent after blocking
    DatagramsUnblocked,
+    /// Received an observation of our external address from the peer.


Explicitly document whether this is necessarily different from the previous one, and the required configuration and peer support.

Ralith · 2024-11-22T21:34:53Z

quinn-proto/src/connection/mod.rs

+                        // include in migration
+                        migration_observed_addr = Some(observed)


Why is this a special case?

quinn-proto/src/frame.rs

quinn-proto/src/address_discovery.rs

djc · 2024-11-25T09:57:44Z

quinn-proto/src/connection/mod.rs

+                    {
+                        return Err(TransportError::PROTOCOL_VIOLATION(
+                            "received OBSERVED_ADDRESS frame when not negotiated",
+                        ));


Could we yield this error from within should_report() instead? (Maybe also pass in packet.header.space()?)

djc · 2024-11-25T10:01:11Z

quinn-proto/src/connection/mod.rs

+        new_path.last_observed_addr_report = self.path.last_observed_addr_report.clone();
+        if let Some(report) = observed_addr {
+            if let Some(updated) = new_path.update_observed_addr_report(report) {
+                self.events.push_back(Event::ObservedAddr(updated));
+            }
+        }


Can we push all/most of this logic into a method on PathData?

quinn-proto/src/frame.rs

quinn-proto/src/transport_parameters.rs

djc · 2024-11-25T10:08:51Z

quinn/examples/client.rs

+    let mut external_addresses = conn.observed_external_addr();
+    tokio::spawn(async move {
+        loop {
+            if let Some(new_addr) = *external_addresses.borrow_and_update() {
+                info!(%new_addr, "new external address report");
+            }
+            if external_addresses.changed().await.is_err() {
+                break;
+            }
+        }
+    });


I don't think this is widely used enough that it merits real estate in these examples.

djc · 2024-11-25T10:10:25Z

quinn/src/connection.rs

@@ -636,6 +636,12 @@ impl Connection {
        // May need to send MAX_STREAMS to make progress
        conn.wake();
    }
+
+    /// Track changed on our external address as reported by the peer.
+    pub fn observed_external_addr(&self) -> watch::Receiver<Option<SocketAddr>> {


Not sure about exposing the watch::Receiver as part of the public API, that seems like it's leaking a bunch of implementation details?

gretchenfrage

Looks promising so far!

gretchenfrage · 2025-01-04T00:32:46Z

quinn-proto/src/connection/mod.rs

@@ -222,6 +222,12 @@ pub struct Connection {
    /// no outgoing application data.
    app_limited: bool,

+    //
+    // ObservedAddr


Probably Address discovery would be a better section divider comment

gretchenfrage · 2025-01-04T00:41:15Z

quinn-proto/src/connection/mod.rs

@@ -3058,6 +3103,53 @@ impl Connection {
                self.stats.frame_tx.handshake_done.saturating_add(1);
        }

+        // OBSERVED_ADDR
+        let mut send_observed_address =


It looks like the only reason for this to be a lambda is because of the short-circuit. I think this can be refactored from:

let do_thing = |a, b, c, d, e| { if (!should_do_thing) { return; } do_stuff.... } do_thing(a, b, c, d, e);

To just:

if (should_do_thing) { do_stuff... }

It would be equally indented.

gretchenfrage · 2025-01-04T00:45:49Z

quinn-proto/src/connection/paths.rs

@@ -37,6 +40,11 @@ pub(super) struct PathData {
    /// Used in persistent congestion determination.
    pub(super) first_packet_after_rtt_sample: Option<(SpaceId, u64)>,
    pub(super) in_flight: InFlight,
+    /// Whether this path has had it's remote address reported back to the peer. This only happens
+    /// if both peers agree to so based on their transport parameters.
+    pub(super) observed_addr_sent: bool,


Could we avoid tracking this extra state by queueing a frame immediately whenever we change the path?

As an example, I am implementing similar "send frames X constant number of times per path" in #1912 with logic here / here / here

gretchenfrage · 2025-01-04T01:10:47Z

quinn-proto/src/connection/paths.rs

+        &mut self,
+        observed: ObservedAddr,
+    ) -> Option<SocketAddr> {
+        match self.last_observed_addr_report.as_mut() {


I think this could be shorted a lot. ObservedAddr is copy. Maybe something like this?

let prev = self.last_observed_addr_report; if prev.map(|prev| prev.seq_no < observed.seq_no).unwrap_or(true) { self.last_observed_addr_report = Some(observed); } if Some(observed.socket_addr()) != prev.map(|prev| prev.socket_addr()) { Some(observed.socket_addr()) } else { None }

It could be golfed more but something like the above perhaps is at a nice spot for readability.

gretchenfrage · 2025-01-04T01:15:38Z

quinn-proto/src/frame.rs

+            IpAddr::V6(ipv6_addr) => {
+                buf.write(ipv6_addr);
+            }
+        }


Nit: Maybe just this?

match self.ip { IpAddr::V4(ipv4_addr) => buf.write(ipv4_addr), IpAddr::V6(ipv6_addr) => buf.write(ipv6_addr), }

gretchenfrage · 2025-01-04T01:16:42Z

quinn-proto/src/frame.rs

+        Ok(Self { seq_no, ip, port })
+    }
+
+    /// Gives the [`SocketAddr`] reported in the frame.


This project's convention is to have the first line of a doc comment not end with a period

gretchenfrage · 2025-01-04T01:20:33Z

quinn-proto/src/address_discovery.rs

+    /// Whether this peer should report observed addresses to the other peer.
+    pub(crate) fn should_report(&self, other: &Self) -> bool {
+        self.is_reporter() && other.receives_reports()
+    }


I think maybe this should just be inlined into its call sites, I don't think the abstraction is helping

quinn-proto/src/address_discovery.rs

divagant-martian force-pushed the quic-address-discovery branch from 276c164 to b0f0a30 Compare November 14, 2024 16:24

divagant-martian force-pushed the quic-address-discovery branch from eaa83bc to 5c0e633 Compare November 21, 2024 17:26

divagant-martian marked this pull request as ready for review November 21, 2024 17:26

divagant-martian force-pushed the quic-address-discovery branch from 5c0e633 to 60b2e57 Compare November 22, 2024 15:33

Ralith requested changes Nov 22, 2024

View reviewed changes

djc reviewed Nov 25, 2024

View reviewed changes

divagant-martian added 4 commits January 3, 2025 12:15

Define address discovery roles present in the extension negotiation

99ae117

add ObserevedAdd frame and frame types

e2d3f2f

add transport parameter for address discovery

2881b9c

Add saturating add for Varint

b62a16e

divagant-martian force-pushed the quic-address-discovery branch from 60b2e57 to 903743a Compare January 3, 2025 17:18

divagant-martian requested a review from gretchenfrage as a code owner January 3, 2025 17:18

divagant-martian force-pushed the quic-address-discovery branch from 903743a to db19d8b Compare January 3, 2025 17:21

gretchenfrage requested changes Jan 4, 2025

View reviewed changes

divagant-martian added 4 commits January 8, 2025 10:11

Add transmission and retransmission logic

cddc7cd

add tests

7732a8a

use in example

5f0214b

move address discovery role from an enum to a simplified struct

6d67b90

divagant-martian force-pushed the quic-address-discovery branch from db19d8b to 6d67b90 Compare January 8, 2025 16:25

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: QUIC Address discovery extension #2043

feat: QUIC Address discovery extension #2043

divagant-martian commented Nov 14, 2024 •

edited

Loading

Ralith left a comment

Ralith Nov 22, 2024

djc Nov 25, 2024

Ralith Nov 22, 2024

Ralith Nov 22, 2024

Ralith Nov 22, 2024

djc Nov 25, 2024

djc Nov 25, 2024

djc Nov 25, 2024

djc Nov 25, 2024

gretchenfrage left a comment

gretchenfrage Jan 4, 2025

gretchenfrage Jan 4, 2025

gretchenfrage Jan 4, 2025

gretchenfrage Jan 4, 2025

gretchenfrage Jan 4, 2025

gretchenfrage Jan 4, 2025

gretchenfrage Jan 4, 2025

		// include in migration
		migration_observed_addr = Some(observed)

feat: QUIC Address discovery extension #2043

Are you sure you want to change the base?

feat: QUIC Address discovery extension #2043

Conversation

divagant-martian commented Nov 14, 2024 • edited Loading

Ralith left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gretchenfrage left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

divagant-martian commented Nov 14, 2024 •

edited

Loading